* feat(xxxLikeOp): ones_like and zeros_like use user op

* fix(Optim): fix learning rate device error bug

* style(*): format codes

* style(*): use int instead of np.int

* test(Optim): add optimizer gpu test (#4908)

Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>